A Wells Fargo branchReuters/Mario Anzuoni

In a short section of its most recent quarterly report, Wells Fargo revealed that for more than five years, beginning in April 2010, the company had made “an automated miscalculation” that had dropped 625 mortgage holders below a threshold where they could receive a loan modification. Four hundred of these people subsequently had their homes foreclosed on. Wells Fargo finally caught the error in October 2015.

Coverage of the problem described it as “an error,” or even “a computer glitch.” The description in the Securities and Exchange Commission filing, meanwhile, raises as many questions as it answers: “This error in the modification tool caused an automated miscalculation of attorneys’ fees that were included for purposes of determining whether a customer qualified for a mortgage loan modification.”

As a result, Senators Elizabeth Warren and Brian Schatz sent Wells Fargo executives a long list of questions demanding to know the details of the case. In sum, their queries add up to an exasperated How could this have happened?

We know this much: Wells Fargo created this software itself, according to Tom Goyda, a spokesperson for the company. “It was a tool that we developed,” he told me, “so the programming error was made internally.”

In most loan modifications, the debtor is not actually getting out of paying back any of the principal. Instead, the bank rolls the past-due principal and interest into a larger loan. The bank (or “servicer,” in industry terms) can also roll a bunch of other things like insurance premiums and fees into the new loan balance. Then, sometimes with government money as an incentive, the bank agrees to lower the interest rate and extend the term of the loan to generate a smaller monthly payment. The target amount for that payment, under the main federal program, is 31 percent of the borrower’s gross income. The new loan also has to pass another test for calculating whether the bank is likely to make roughly the same amount of money on the new loan as the old. If it does, and the debt-to-income ratio is right, perhaps the borrower can get a loan modification. If not, then it’s a no-go.

These calculations were made in tremendous numbers during the foreclosure crisis (which was precipitated, in part, by the bad loans that banks targeted to largely black and brown neighborhoods, and for which Wells Fargo was fined $175 million.) Under the most important government-backed loan-modification scheme, HAMP, or the Home Affordable Modification Program, Wells Fargo alone received 1.6 million applications from desperate people.

For these borrowers, the “automated-decisioning tool” that Wells Fargo built took all of the variables from individual loans and borrowers, and combined them with various constants—including, crucially, that incorrect calculation for attorneys’ fees—and came back with a thumbs-up, thumbs-down determination of modification eligibility. But because of that miscalculation, the automated tool gave a thumbs-down to people who were right on the borderline of getting a modification and who should have gotten a thumbs-up.

Only the decision—not its actual calculations—was rolled forward to other parts of the bank, Goyda said, so no one saw the erroneous attorney-fee number. That’s also why, as Wells Fargo disclosed, customers were not actually overcharged; the standalone tool made the decision, but the actual loan terms were dealt with by other means.

Wells Fargo approved 28 percent of modification requests, a little below the average for the four biggest servicers. The number of people affected by the attorney-fee error added up to 0.0386 percent of that HAMP pool.

Set against the massive scale of the Great Recession—9 million jobs lost, 9 million homes lost—this is the kind of small error that could seem insignificant. But this is hundreds of lives irrevocably changed, with all the ripples outward. And there’s still a lot we don’t know, as indicated by the senators’ letter full of questions.

For example, why did it take three years for Wells Fargo to report the problem after it had found the error?

“It was identified and corrected at the time, but this particular impact wasn’t something that anybody thought to look for,” Goyda told me. “Subsequently, we decided that was something we needed to do, and responded quickly once we discovered that these impacts existed.”

And why has the bank only set aside $8 million in projected costs to redress these problems, which works out to just $12,800 a borrower? And is this really the only error there is among that huge mass of mortgage modifications and foreclosures?

“Historically, miscalculations in [loan] reviews are not uncommon. They normally show up in income calculations and appraisal figures,” Alys Cohen, an attorney at the National Consumer Law Center, told me. “What’s different here—but probably not unique—was that it was an input to the software that was uniform for everybody. And the question really is: How many more of those existed both at Wells Fargo and at other companies, and what work has anyone done to identify those problems and redress the losses to homeowners?”

And finally, Wells Fargo didn’t do this off in some private corner. It was often working with government money to revamp these loans. Fannie Mae administered the program, Freddie Mac provided oversight, and various regulators could have been expected to do due diligence as well. “Where were the regulators in all of this?” Cohen asked.

One thing that’s clear from the debacle is that Wells Fargo’s in-house software—whether it was just an Excel spreadsheet or something more sophisticated—meets the three criteria of a “weapon of math destruction,” as the author Cathy O’Neil memorably termed this kind of tool. It operated with opacity at scale and generated real damage. And if there were “a Mueller probe but for all of US banking during the housing bubble and crash,” as MSNBC’s Chris Hayes recently suggested, how many other financial bombs are still lurking in the wreckage of the Great Recession?

We want to hear what you think about this article. Submit a letter to the editor or write to letters@theatlantic.com.